KK reports on the Tol vs Curry fight. Tol is complaining that Curry is doing her usual: posting about septic junk and then saying "oh but I'm just asking". Tol may have long hair but, unlike Curry, he isn't a bozo, or irresponsible (he did call me rude things in an email once, but I forgive him).
[Update: incidentally, there is an interesting exchange between KK and RP Jr (!) in the comments:
KK> Do you assign lousy, error-riddled textbooks for your class to read?
RP> Yes, absolutely. The Skeptical Environmentalist was a core reading...
It is interesting only because that was a silly question from KK, and a failure-to-think response from RP (or rather, a point-scoring response). The point is, within a managed class structure with someone guiding the discussion, it is fine to discuss flawed texts, for the reason given: it encourages critical thinking. That wasn't what Curry was doing. There was no guidance at the start, nor does she guide the subsequence discussion.
Also, Gavin's point]
[Updated again: this turned in the comments into McShane and Wyner and the "Lasso" method. So I've added a pic of Lasso being a bit crap, from Gavin et al.'s reply.]
Question - Tol is on the "academic advisory council" of the Global Warming Policy Foundation, those good-faith(not) folks whose banner graphic shows global(?) temps for the too-short period 2001-2011, with a nice nonstatistical flat line drawn through the upper points.
Why would a reputable academic allow himself to listed as an advisor for a group that's not following intellectually honest (or competent) advice, with world-damaging consequences?
Does this mean that we heuristic-reliant laymen should take Tol's critique of the papers with extra salt?
(caveat, I have not read the JC posts or plethoras of comments; just skimmed somewhat. Obese writing puts me to sleep too.)
Steve, the "leads me to believe" part of your comment strikes me as being out of line; regardless of the provocation.
Why be nice to Richard, is he immune to his own medicine?
But he likes the McShane and Wyner paper. Which I still find a bit uncomfortable.
That being said, Judith did give him a guest post (good on her), in which he dissects the "septic junk" in question quite thoroughly.
[Yes, though that doesn't make her any more clueful. McS+W: I don't think there is any need to assume Tol knows that stuff; I don't think he claims to -W]
Re: McShane & Wyner
Tol argued with Eduardo Zorita about the merits of the paper when it appeared.
He has also referenced Beenstock & Reingewertz in the thread, saying:
'There seems to be a fair amount of agreement that the atmospheric greenhouse gases are a statistically adequate explanation of the observed warming (with one important exception: economics.huji.ac.il/facultye/beenstock/Nature_Paper091209.pdf)'
As far as I can tell this paper hasn't actually been published anywhere, the link goes to a pre-review copy that has been floating around the Internet for a couple of years.
[Yes. It is hard to see why anyone would think that was an "important" exception; it looks like the usual nonsense -W]
He seems to trust statistical arguments over physical ones. Perhaps understandable given his background.
Re: Judith Curry
The Ludecke et al. papers aren't exactly any worse than stuff she usually posts on her blog so it's slightly random that this is what brought a rise out of some people. She's made it quite clear this is how she wants to proceed. I think people need to adjust their expectations to an understanding that anything posted on Climate etc. should, a priori, be allocated no more credibility than anything on WUWT - perhaps that should be her strapline.
[I agree that she should not be taken seriously; I certainly don't -W]
Calico caat and gingham dog.
Both completely ignorable.
Up to now my knowledge of Curry's views has mostly been second-hand, from things like blog comments and interview snippets. So I decided to click through to her "Pause (?)" post and see what she had to say.
I half expected to be outraged, or at least skeptical. But after reading several paragraphs of meandering, squishy prose what I got instead was an urge to have a nice lie down.
Can anyone else figure out what she's trying to say? If so, what's the magic decoder key? Does she always write like this, or did I catch her on a bad day?
I suspect that she's trying to balance her new role as Patron Saint of Lost Souls against her credibility in the scientific community. So what we get is this verbose circumlocution where everything is loaded down with qualifiers and vague conditions. Another possibility is that she has always written this way, and no one noticed until she had a venue where she could post her thoughts with no editing. (She wouldn't be the first accomplished scientist who has trouble writing clearly.) Or it could be that even she doesn't know what she's trying to say; in other words, unclear writing is the product of unclear thinking.
I'd be interested in comments from experienced Curry-watchers.
[I think you're right that she is trying to strike a balance. So when she puts up posts pointing to other people's nonsense she is fairly careful not to actively endorse it. OTOH stuff like the "pause" post is her very own, and perhaps more egregious is her "Italian flag" uncertainty analysis, which JA amongst others has ripped to shreds (see links herein). She, like many people, have ideas in their head about how things work; those ideas are half-thought-out and in many cases incoherent; but then she writes them down whereas most people have the sense to keep quiet.
Also, she is enjoying her days in the sun. She knows that in order to stay there she needs a constant stream of posts to feed people with. But she doesn't have enough good, sensible material to feed that stream - though to be frank, I've seen very little on her blog that could be considered good, or even not-bad (not in the English sense, of course). The correct choice would be to post less -W]
Where does he say he likes McS+W?
The issue with the McShane and Wyner paper is that it was mathematically not wrong. It contained a load of other crap ("MBH did X and Y"...errr, no they didn't), and there was the usual sophistry with using certain statistical methods while not showing it is any better than previous choices, but that's not something Tol is going to complain about.
John: he repeatedly mentions McS+W in the two threads at JC's.
A skim of those threads is quite entertaining. The peanut gallery there are repeatedly attacking him as a warmist, and demanding he defend stuff which he doesn't hold (such as the hockey stick). Because he's on The Team. The gallery has no time for nuance.
Can anyone else figure out what she's trying to say?
As little as possible, I suspect.
the usual sophistry with using certain statistical methods while not showing it is any better than previous choices,
I think that's precisely the problem, because these methods are the contribution.
Shorter McS&W: "We find proxies aren't better than noise when testing them on small-period interpolation, therefore we conclude that the proxies really have no signal, and blatantly ignore all the work that seems to show they really are better than noise at long-period, open-ended prediction."
Back to topic: the authors of the "septic junk" in question have responded to Tol. It's... entertaining.
Is there a discussion over here somewhere about Curry's "Disinformation and pseudo critical thinking" post? I skimmed it, & didn't noticed that it contained one of the most effective "intending to mislead" tactics, namely focusing on diversionary stuff that obscures the larger picture. But if this isn't disinformation, what _is_ the larger term for what it is?
(i.e., for uttering only true information but nonetheless intending to mislead)
I know I know it, but am drawing a brain blank.
[I'm doing my best to ignore Curry. Yes, I know this post isn't exactly a shining example of that, but I'm using Tol as my excuse, because he is more interesting -W]
Here's one: Tol on McS-W:
Thanks, all. It is rare for me to visit there, so I hadn't seen those.
I did go over and look, running across this, which actually encapsulates the whole point:
"Joe, the thing to learn is that this is complex and controversial, as the frontiers of science always are. If you are not an expert on this stuff, as I am not, then there is little to learn except that the experts disagree. On the other hand, I have learned a lot about what I do not understand."
[I agree that she should not be taken seriously; I certainly don't -W]
I don't agree that she shouldn't be taken seriously but Curry's gone in a direction with her blog that makes the fact that the host is a climate scientist almost an irrelevant curiosity (though perhaps the attraction of the site is that curiosity).
People should adjust their expectations accordingly, rather than expecting Curry to adjust her ways. It would probably be better if she made her mission clearer though, lest some unsuspecting journalist pop by and assume the nonsense currently under discussion holds some genuine significance.
Speaking of nonsense under discussion there, I did submit a short piece to her last week - curiosity, you see - about the accuracy of the 1990 FAR projections, asking the 'denizens' what would have been their predictions based on the little warming seen at that point in the 1990 global temperature graph. According to her accounts she must have received that at about the same time she received the Ludecke papers, but nothing so far.
Paul S fan here! Hope she publishes it. Your comments are always excellent: concise and easy to understand, and you clearly keep up with the subject. I think what you are saying about Juday, Juday, Juday is pretty much right on.
I thought one of the most interesting comments about McS-W was made by MattStat. Did you notice that?
A while back I went over Judy's pubs and found almost nothing in the way of first/sole-author papers (some exceptions for data collection campaigns IIRC, although those had the usual cast-of-thousands author list). That, combined with the recent sea ice paper with the novelty problem (something Judy should have known about because she was working in the field at the time the prior paper was published -- heck, *I* knew about it), leads me to believe that her scientific chops were thin to begin with and that she's possibly showing the signs of aging (both of which would also help explain the writing style). Also, she and Peter have a product to sell (via the tropical met forecasting business they set up around the time Judy started her blog), and maintaining a high public profile (not just the blog, but becoming a go-to contrarian for the science media) is probably helpful for that.
Paul S, that's a brilliant idea; I hope Dr. Curry agrees.
Back to Richard Tol, why would an academic be willing to lend his name (as Academic Advisory Council member) to an org whose banner image showed a short-term graph to obscure a long-term and extremely worrisome trend?
[He may not regard himself as "lending his name", though most will think he is. He is on their advisory council, but we don't know what advice he is offering -W]
You seem to have missed my follow up comment to RPJ:
"Iâm also guessing that you at least (somewhat) guide your students through their reading of The Skeptical Environmentalist."
[Yes, I missed that. But you're missing my point: that your original question was a poor one, in that it confused the two situations -W]
The Ludecke et al. papers aren't exactly any worse than stuff she usually posts on her blog so it's slightly random that this is what brought a rise out of some people.
I might have made it worse through my righteous indignated ranting (sorry if I've embarrassed anyone).
I normally never go to Curry's blog because of the vagueness, lack of transparency/consistency and WUWTesque comment section. However, the Daily Mail-pause-thingy caught my attention (via Tamino), and I was reading the comments to see what Curry would say and do.
In an offhand remark she mentioned those EIKE papers. I immediately told her that EIKE was not to be trusted, and Fred Moolten straight away spotted a few problems. That she posted the EIKE disinformation crap anyway the following day without any commentary (similar to the way Morano does it actually), got my hackles up and made me call her out on it. Surprisingly, Tol did the same.
This was the sign for Curry's drooling watchdogs to jump on me and Richard Tol and another 500+ comment muddle was born. In between Curry called me a CAGW ideologue, which doesn't offend me, but which says something. Not that we didn't deduce it already from her uncritical swallowing of the McIntyre narrative.
She's given EIKE a huge platform (LÃ¼decke gets to call her 'Judy', so I'm sure a lot of mutual admiring e-mails have been sent back and forth), a lot of time is wasted and no one will learn anything, either because they already knew it (99% of what EIKE puts forth is disinformation propaganda to delay policy) or they don't want to learn.
And if you say something about that you are a gatekeeper and CAGW ideologue. But how much crap needs to be reswallowed before we can discard a source as untrustworthy? When can we ignore someone, no matter how hard he/she shouts? Unfortunately Curry won't really answer that question.
I'm following the discussion on BEST because I'm thinking of updating Tol and de Vos (1997): http://dx.doi.org/10.1023/A:1005390515242
I paid attention to Climate Etc in the beginning but quickly lost interest. The link to BEST brought me to the Ludeke papers. I was shocked that she had put them up. I did not know that she had made a habit of doing such things.
McShane and Wyner may be badly written, but it is by far the most advanced statistical analysis of the proxy record. While many see proxies as a physical or biological problem, I view it as a statistical problem: What information is shared between the proxy and the instrumental record? And how can that information be used to infrapolate the instrumental record?
[I think you may be imposing your own biases, to some degree: you have a background in the stats, so that is what you focus on. Treating climate purely as a stats type problem has a long record of working very badly.
That said, I haven't read M+W in detail, so can't vouch for its flaws as reported in e.g. http://deepclimate.org/2010/08/19/mcshane-and-wyner-2010/ or http://www.realclimate.org/index.php/archives/2010/12/responses-to-mcsh… . I'm curious: do you claim to have read M+W in sufficient detail to be able to dismiss the criticisms? -W]
Those two commentaries, as well as Zorita's referred to earlier, just demonstrate that the authors did not understand what McShane and Wyner did. LASSO, in particular, baffles them. LASSO is so far ahead of the usual stuff in paleoclimatology that they fall to grasp even the basic intuition.
[You can assert that, if you like, but without some backing it won't convince anyone who doesn't trust your personal authority. Quite possibly you don't care, but if you do, it might be worth writing down somewhere. Also, you didn't answer my question -W]
Yes, I studied McS+W in detail.
Wikipedia has a succinct description of LASSO: http://en.wikipedia.org/wiki/Lasso_%28statistics%29#LASSO_method
It is clear that Mann & Schmidt, who worry about over-fitting, did not even get to the fifth word on Wikipedia. DeepClimate does not realize that LASSO generalizes his preferred methods. Zorita does not understand encompassing.
McShane has a PhD in statistics. Wyner is a professor of statistics. They're sound.
[Wegman is also a prof of statistics, so just that doesn't make you sound :-). But maybe M+W are; that loops round to my initial point.
Thanks for the pointer to the description of LASSO. OK, so it is an L1 constraint on the L2 fitting. It isn't clear to me why that is what one wants (but please note I'm not claiming to have listened to the details of the debate over methods in detail) -W]
You impose an L1 constraint if you worry about outliers.
[I wrote "That doesn't make sense to me. You want to try to ignore outliers; L1 will force you to follow them, no?" before realising that I'd confused L1 and L-inf. So, at this point I think I should admit that I haven't thought / read all this through. If you pointed me to a coherent explanation of why M+W was a good idea, I might read it -W]
It is infinitely amusing to watch Richard act as a statistics totalitarian. Obviously IOKIYART.
[Be nice. RT is a guest, and a welcome one -W]
Can you comment on the decision to use an interpolation task for validation, and the apparent disregard of previous results based on half-record reconstruction? I know I'm not the only one troubled by this but maybe that's just a misunderstanding.
"It is clear that Mann & Schmidt, who worry about over-fitting, did not even get to the fifth word on Wikipedia. DeepClimate does not realize that LASSO generalizes his preferred methods. Zorita does not understand encompassing.
McShane has a PhD in statistics. Wyner is a professor of statistics. They're sound."
You may be correct in that, but why all the peripheral snark and misrepresentation of MBH? Let's improve the stats, but that doesn't mean ignorance or lies on their part is required for what they are purportedly doing.
Richard, is it a "sound" starting point to make false claims about a paper you are criticizing (as Eduardo discussed in his criticism)?
Toto (25), Marco (27)
I don't understand the question.
The first half of the McS&W paper was a big mistake. I don't understand why the referees and editors let them do that. I don't know why they did include all that stuff in the first place, because it gives a lot of people the perfect excuse to ignore the excellent second half of the paper.
[I'm glad you said that: a fine display of avoiding tribalism, in the argot of the blogs. I agree, including the "excuse" bit, though not having read the thing I don't endorse the "excellent" bit -W]
There were two comments to McS+W that discussed how poor a method LASSO was for the paleo-climate problem:
Alexei Kaplan also had interesting things to say:
Contrary to Tol's claims, the Schmidt et al response (http://arxiv.org/abs/1105.2145v1) did not discuss the details of LASSO, other than to demonstrate that the method, as implemented by McS+W, performed abysmally on pseudo-proxy data taken from climate model runs:
Not the first time Richard Tol makes a strong claim and then never gets around to providing evidence.
About LASSO, you don't actually have to know much about its soul life if you have running code which tells you how piss-poorly the method performs on verification -- also with simulated proxies, where you know what went in and what is supposed to come out. I mean, isn't this what verification is for?
Also someone, not me I hasten to add, might ask why, if LASSO is the best thing since sliced bread, do M&W bother to produce a bundle of reconstructions using a more traditional LS approach, using different regularization parameter values? I found that of that bundle, half lie bang on top of the original Mann et al 2008 (in spite of an error in their processing chain as pointed out by Gavin), whereas the other half are all over the place, apparently due to ineffective regularization, as also borne out by their verification scores.
I played around with their code for a while, until I let it rest as Gavin and Mike were doing a so much better job of it. In the thread tradition of saying something nice and humoring William, M&W's R code was well written and easy to work with.
Richard, in a way you already half-answered my question by answering Harry.
It is a sad reflection on the state of climate research that you respond the way you did to McShane and Wyner, or for that matter to McIntyre and McKittrick.
As you probably know, I was a professor in the geosciences department at Hamburg U for 7 years; have published a number of papers in statistical climatology; and used to follow that literature quite closely.
If you look in your heart of hearts, you know that only a few geoscientists have had a rigorous training in statistics and that you are not among those few.
It is unfortunate that MsS&W used such an aggressive tone. They used a method that is not perfect, no method ever is, but that is vastly superior to the standard techniques in paleoclimatology. In their rejoinder, they again explain, remarkably patiently, why that is the case, and provide another battery of tests to prove their point.
You should really take advantage of the expertise on offer.
[We've had Wegman, I'm afraid, so "I'm a prof of stats, trust me" doesn't work. You've asserted that M+W is "vastly superior to the standard techniques" but haven't provided any evidence to support that view. OTOH, Gavin has provided a nice link to a paper which contains a nice pic which says that LASSO doesn't work very well on these problems. You're welcome to provide a link to something that provides evidence to the contrary -W]
Gavin omitted the link to the rejoinder by McS&W: http://arxiv.org/abs/1105.2433
[Thanks. I'll look. So far I've got to Among other things, the RegEM EIV ï¬tting procedure cannot be executed by a straightfoward function call as is typical for statistical code libraries. Rather, the archives consist of a large number of ï¬les layered on top of one another and, despite a major eï¬ort on our part, we were unable to replicate published results within the publication time constraints of this rejoinder which is pathetic whinging; I hope they get better later -W]
Your line of reasoning is peculiar: Because Wegman has been accused of plagiarism, which casts doubts on the authorship rather than on the quality of his work, geoscientists are better at statistics than statisticians.
[No. Plagiarism is only what he has been convicted of. He is guilty, too, of incompetence. But we probably don't need to go down that route, because you've misrepresented my argument. All I'm concluding is that "prof of stats" is no guarantee of competence or quality. Which is why your apparent argument-from-similar-authority won't do. "geoscientists are better at statistics than statisticians" is pure putting words in my mouth, which is unworthy: you can do much better than that -W]
[Further to M+W rebuttal: their reply to the point of fig 1 isn't convincing. This really comes back to the point where we started (again). They seem, like you, to think that Pure Statistical Thought without any understanding of the data will do fine' I don't believe that. As to fig 2 - the bit about LASSO being poor - they end up with "There are no statistically signiï¬cant or even practically important diï¬erences between our modelâs reconstruction and that of RegEM". Which might even be true - but it hardly supports your assertion that it is vastly superior -W]
You are the one who raised Wegman (twice), so forgive me for assuming that you thought he had anything to do with this discussion.
[Certainly I raised him, in your response to your one-of-my-guys-is-a-stats-prof-so-must-be-OK argument from authority -W]
The finding that the SMR procedure has not been consolidated in a single script is rather worrying as it makes their work irreproducible and intransparent,and it indicates a lapse in quality control.
[Err, no it doesn't. That is a simple logic fail on your part. And M+W don't even say that, they say single function call, which is different. C'mon: this isn't the heart of the matter, stop arguing trivia -W]
"The finding that the SMR procedure has not been consolidated in a single script is rather worrying"
It is neither worrying, nor true. Code is available here:
Do try to keep up.
Among other things, the RegEM EIV ï¬tting procedure cannot be executed by a straightfoward function call as is typical for statistical code libraries. Rather, the archives consist of a large number of ï¬les layered on top of one another and, despite a major effort on our part, we were unable to replicate published results within the publication time constraints of this rejoinder
Oh, wow. That's hilarious. All that actually needs changing to make the code work is a few hardcoded file paths, and that's stated explicitly in the readme. That and a couple obvious errors in one module which can be fixed with a simple find/replace. A few minutes' work for anyone who knows what they're doing.
Having spent a few weeks on this last year (Appendix A.12 in SSWR, and having read the various critiques, I figured this topic was done.
However, maybe people might wish to comment on the professional PR campaign:
[The odd thing is, when you get down to it finally, is that there isn't really much to M+W. Tol claims their method is far better, but they themselves claim their method is only about as good as previous. The only excitement was about the early part of the record, which they had as being warmer; but they don't really defend themselves from the criticism that that bit is wrong. Which leaves only the vitriol in the first half, which everyone agrees is an embarassment that should never have survived a competent editor -W]
Yes to all that, but I am curious what people think of the Kellogg press release as a portrayal of the strenght of M-W's results.
[Well, its a press release, innit guv? You don't take those seriously. Or at least, I don't -W]
Well, of course I don't take press releases seriously, although I've certainly helped write a lot [but mine were honest!]
But let me enumerate a set of characteristics:
1) Some people with PhDs in statistics
a) With demonstrated ignorance of paleoclimate, climate science, and sometimes even basic physics
b) And no serious history of work in the field
c) But with disdain for climate scientists
2) Spend a *lot* of time writing an attack on the hockey stick and the scientists who did it.
3) The work starts with substantive plagiarism and falsification/fabrication, imbued with simply-ludicrous errors, in an attempt to claim credibility in the subject.
4) The statistics sounds good to the general public, but turns out to be at best poor, and at worst, cherry-picked in ways to amplify doubt.
5) The work is either announced in the Wall Street Journal or unusually covered by it.
6) Despite strong criticisms, the work gets pushed by a powerful PR campaign.
7) Steve McIntyre gets thanked for help.
Which of those apply to Wegman Report? Which to M+W (including Rejoinder)?
I leave a hypothetical question: is there any difference between doing science and a PR campaign?
Thanks JCH & Anna,
I haven't heard anything from Judith, apart from a quick reply 2 weeks ago saying she was busy. My guess is she won't post it at the momemt because it doesn't fit in with her current narrative arc and she'll have forgotten about it by the time she moves on.
I don't think 3 and 4 are appropriate to list in this way - in what appears to be an allegation against McShane and Wyner.
5 and 6 are not necessarily under the control of the authors.
> My guess is she won't post it at the moment because it doesn't fit in with her current narrative arc
Some encouragement might be helpful.
Paul S., if you're out there, please resend your "guest post" missive; it was not received.
(...or at least not found, by JC)